No bot is perfect.
Projections play a big part in fantasy baseball prognostications, from guiding rankings to assisting in trade evaluation to estimating individual game matchups. They do not always, however, tell a complete story about what is yet to come. Projections generally cannot account for the human element of the game, especially dramatic, unexpected changes from what showed up in yesterday's stats.
Among those changes can include injuries, shifts in a hitter's bat speed or pitcher's velocity (which can sometimes be tied to injuries), or subtle tweaks players make to their swings, batting stances, pitching deliveries or even individual pitch selection. As those things deviate from players' historical stats (which fuel the projections), it's practically impossible for any projection set to entirely account for them in the formulas.
It's precisely those changes, though, which are informational gems we constantly seek to gain an edge in our fantasy leagues. They also present instances where we need to break from the projections in our own decision-making process.
Still, as a projections-oriented analyst, I'm frequently wondering to what degree we can account for skills changes in our projections models. It's something I'm constantly tweaking with the Forecaster formulas, especially with the rising number of advanced analytics available.
Breaking out in Pittsburgh
This past week brought up a great example, with the breakout candidacy of Pittsburgh Pirates RHP Mitch Keller. Entering play on Wednesday, he's the No. 3 pitcher in terms of fantasy points (and the No. 5 SP on the Player Rater) for the season.
The Forecaster formulas don't seem to like Keller, generally granting him an eight-point projection for his recent starts, a "good, but not great" number that lands him in a similar range to Marcus Stroman, Jack Flaherty and Andrew Heaney. That's in part because, as with most any projection model out there, there's no recentcy bias, and the "hot hand" isn't going to garner a big advantage from a recent handful of excellent outings. Two years of historical data remains more important in determining a pitcher's future.
In Keller's case, however, I think the projections get it mostly wrong, but that demands as much from us to believe in more than what the mere numbers say as it fuels the tweaks I continue to make to the formulas. After all, isn't the fun of this game to find the numbers we don't believe in -- players we feel will exceed what the projections say -- and reap the rewards of those against-the-grain calls?
Digging deep on Keller, his breakthrough thus far has been fueled by a number of factors, but perhaps the most notable one has been the return of his cutter, which he used sparingly in 2020 and 2021 but upon which he now relies nearly one-quarter of the time (24.4%, to be exact). Right-handers typically use the cutter to neutralize left-handed bats, and Keller has been no different, with his performance against them being as different as night and day:
Keller vs. lefties through 2022: .307/.390/.445 allowed, 16.9% strikeout rate
Keller vs. lefties in 2023 so far: .194/.256/.315 allowed, 32.5% strikeout rate
Additionally, Keller has benefited from a change to his breaking pitches that occurred in the early-to-middle stages of last season -- most interestingly our pitch-tracking tool changing the classification of his slider to a sweeper due to the dramatic shift in break on the pitch. Consider that through the first 46 starts of his big-league career, Keller's ERA was 6.12, his WHIP sat at 1.70 and batters had hit .306 against him. That ranked him third-worst, worst, and third-worst among the 150 pitchers who had made at least 30 starts over a similar time span. He had four quality starts, ultimately earning him a brief demotion to the bullpen last May.
In the 28 starts since the first outing in which he demonstrated this sizable change in movement in his sweeper, last June 19, Keller has produced 18 quality starts, a 2.97 ERA, a 1.23 WHIP and a .244 BAA. Bear in mind that 19 of those 28 starts, too, occurred last year, before he reintroduced the cutter into his repertoire, lending further legitimacy to 2023's to-date returns (as does his 2.90 Statcast expected ERA).
We, the lot of us fantasy baseball managers, tend to default to "regression" with pitching examples like Keller's, and by all rights he probably shouldn't be expected to have a 2.38 ERA over the remainder of the year. The projections will certainly concur, and even I have him ranked merely as my No. 49 starting pitcher for the season's remaining 4 1/2 months. Nevertheless, with the adjustments he has made, he's the kind of "sell high" candidate and modest-projection type that I'd practically refuse to trade away.
Keller is one of the handful of examples, as we see each year, where the height of the prospective breakthrough is too valuable to trade at anywhere near his current ranking, projection or market-perception price point. It helps, certainly, that he was regarded a near-unanimous top-10 pitching prospect at the time of his big-league debut, meaning that the prospect pedigree is there.
Just keep reminding C-3PO, "Never tell me the odds."
Other 'projections outliers'
Toronto Blue Jays RHP Alek Manoah, a large part of the subject of last week's column, might be the ideal counter-example to Keller's breakthrough, a highly touted preseason pick whose skills have waned to the point that I'm frequently working to push down his projection and ranking and would trade him anywhere I still could.
San Diego Padres LHP Blake Snell is another who resides on that side, a top-30 SP selection in NFBC leagues during the preseason who has shown sharp downturns in average fastball velocity (94.7 mph, down from 95.8 last season), first-pitch strike rate (51.9%, which would represent by far a career worst and well beneath his 59.3% career number) and ability to limit hard contact (career worsts with a 41.7% hard-hit rate allowed and .360 expected wOBA).
Snell has long faced criticism for his inefficiency with pitch counts, but his diminished stuff has me skeptical that he'll deliver consistently productive fantasy numbers. As he's riding a streak of three consecutive quality starts, though, he's an ideal "sell low" candidate.
Almost every indicator is pointing upwards with Minnesota Twins RHP Sonny Gray and, like Keller, he's a pitching chip I wouldn't try to cash in. His velocity is up, albeit slightly (0.6 mph, to be precise), and his curveball is getting 17% more whiffs (the percentage of swings that result in a flat-out miss) this season than last. Gray has also shelved a slider and regularly uses a cutter for the first time in his career, often serving his "go-to pitch" with two strikes. He's another pitcher who was a recent column subject.